Fairness in Algorithmic Decision Making

This article explains how machine learning algorithms work and how they are being used to make critical decisions. Eliminating bias in these algorithms and making them fair to groups and individuals is vitally important. We review a few common definitions of fairness of algorithms, and point a way out of the seeming impasse of choosing between mutually incompatible criteria in some scenarios. We identify two things that algorithmic decision-making systems can learn from human decision-making systems – rules-of-evidence-type limitations on the kinds of data that may be used, and exercising great forbearance in making highly preemptive decisions. Finally, we describe what machine learning models are and describe the need for transparent and accountable models.

By Sampath Kannan^[1]

Machine learning algorithms are being used to classify individuals in increasingly consequential situations. Such algorithms are used for example, in hiring, college admissions, bank loan decisions, and in the criminal justice system for predictive policing and predicting recidivism risk. It has been observed that while algorithmic decision-making may appear prima facie to be more objective than human decision-making, it inherits some of the problems of human decision-making, and presents new ones.

How do machine learning algorithms make decisions? What are their desiderata and what yardsticks do we use to measure whether they are met? What new challenges do algorithms present, compare